RBF neural network parameters optimization based on paddy field algorithm

With regard to the issue of selecting Radial Basis Functions (RBF) neural network center parameters, this paper has introduced the paddy field algorithm (PFA) for its optimization. PFA had stronger global search capacity and higher convergence speed so as to better optimize RBF neural network. In the simulation experiment, this method was applied to approximation and prediction of a typical nonlinear function and compare with PSO (Particle Swarm Optimization) algorithm and the methodology of training by traditional gradient descent algorithm. The experiment showed that all predicted errors were lower than that of PSO predicted results.